Mining　subtopics　and　analyzing　their　sentiment　dynamics　on　weblogs　have　many　applications　in　multiple　domains.　Current　work　pays　little　attention　to　the　combination　of　topics　and　their　sentiment　evolution　simultaneously.　In　this　paper,　we　study　the　problem　of　topic　detection　and　sentiment-topic　temporal　evolution　in　weblogs,　and　propose　a　novel　probabilistic　model　called　topic　sentiment　trend　model　(TSTM).　With　the　model,　we　can　integrate　the　topic　with　sentiment,　and　analyze　the　temporal　trend　of　the　sentiment-topic.　Experiments　on　two　Chinese　weblog　datasets　show　that　our　approach　is　effective　in　modeling　the　topic　facets　and　extracting　their　sentiment　dynamics.　©　2012　IEEE.